Parametric design and intelligent optimization method of stretch-driven soft gripper

Abstract

A parameterized design and intelligent optimization method of a stretch-driven soft gripper, comprising: a soft gripper including a core gripping part and two side stretching parts; the two side stretching parts are at both ends of the core gripping part; parameters of each part of the soft gripper are design variables; a combination of the design variables is randomly sampled in a specified range to obtain a parameter sample; a geometric model of the soft gripper is constructed; a node set displacement and an initial coordinate of the geometric model are obtained; the soft gripper is deformed, and a center line contour point cloud of the deformed model is calculated according to the node set displacement and the initial coordinate; the parameter sample and the center line contour point cloud of the deformed model are combined to construct a design variable and a contour point cloud dataset; a forward proxy model for predicting a target shape contour point cloud of the design variable and a reverse inference model for inferring the design variable according to the target shape contour point cloud are constructed and trained; the target shape is input into the trained reverse inference model to generate optimal design variables of the soft gripper; and the optimal soft structure under the optimal design variables is verified.

Description

Technical Field

[0001] This invention relates to the fields of soft robotics and artificial intelligence, and in particular to a parametric design and intelligent optimization method for a stretch-driven soft gripper. Background Technology

[0002] With the rapid development of industrial automation, intelligent manufacturing, and human-machine collaboration, unprecedented demands have been placed on mechanical gripping systems. Traditional rigid manipulators rely on complex sensing and control mechanisms. While they can achieve precise operation, they struggle to adapt to unstructured environments, lack robustness in grasping fragile or shape-variable objects, and pose challenges to the safety of human-machine interaction. Soft grippers, with their unique material composition and structural design, overcome the limitations of rigid manipulators. Typically composed of flexible materials such as silicone and hydrogel, they possess characteristics such as large deformation capacity, material flexibility, and environmental adaptability, enabling adaptive and non-destructive grasping of objects with complex shapes. Therefore, research on soft grippers is receiving widespread attention from academia and industry, becoming a crucial breakthrough for improving the intelligence and flexibility of end effectors.

[0003] The basic principles of soft grippers can be broadly categorized into three types: obstruction principle, fluid actuation, and smart materials. Existing technologies utilize the obstruction principle to design universal grippers. The principle is that when the flexible gripper adapts to the shape of the target object, negative pressure drives the structure to contract and grasp the object. Existing technologies have designed fluid-driven six-fingered soft grippers. The principle is to achieve bending deformation of the structure through inflation and deflation, thereby completing the grasping and releasing actions. Existing technologies have proposed a pneumatic soft gripper whose gripper length can be adjusted according to the size and shape of the object, improving the grasping speed. Regarding smart materials, existing technologies have proposed a soft gripper based on shape memory alloys. Under voltage actuation, the soft gripper deforms to grasp the object, featuring a large deformation stroke and high output force. In addition, wire-driven soft grippers are also a novel structural design in recent years. Their principle is to change the shape of the flexible device through an external source, typically following a specific circuit path and fixed points.

[0004] In summary, soft grippers possess significant advantages in flexible grasping and complex environments due to their unique material composition and actuation methods. However, the design of soft grippers faces certain challenges. Structural design and actuation methods are strongly correlated, and grasping performance (such as grasping force, response speed, and durability) is highly dependent on multi-dimensional parameters such as geometric configuration and actuation method. Traditional actuation methods mainly include pneumatic and electric drives, and the grasping models are typically multi-finger grasping, lacking novel actuation modes characterized by enveloping grasping. Furthermore, traditional structural design methods rely heavily on engineer experience or trial-and-error experiments, resulting in drawbacks such as long processing time, high cost, and low efficiency. There is also a lack of efficient reverse engineering methods for objects of different shapes. Summary of the Invention

[0005] In view of the above-mentioned deficiencies of the prior art, the present invention provides a parametric design and intelligent optimization method for a stretch-driven soft gripper. It proposes a wrap-around soft gripper prototype based on the mechanical stretch-driven mechanism, constructs a neural network model to realize intelligent optimization design of the soft gripper, and improves gripping efficiency.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A parametric design and intelligent optimization method for a stretch-driven soft gripper includes the following steps:

[0008] S1. Design a soft gripper including a core gripping part and two side stretching parts; the two side stretching parts are located at both ends of the core gripping part; use the parameters of each part of the soft gripper as design variables; randomly sample the combination of design variables within a specified range as parameter samples;

[0009] S2. Construct a geometric model for the soft gripper; obtain the node set displacements and initial coordinates of the geometric model; deform the soft gripper and calculate the centerline contour point cloud of the deformed model based on the node set displacements and initial coordinates;

[0010] S3. Combine parameter samples and deformed model centerline contour point cloud to construct a dataset of design variables and contour point cloud;

[0011] S4. Construct and train a forward surrogate model for predicting the target shape contour point cloud based on design variables and a backward inference model for inferring design variables based on the target shape contour point cloud.

[0012] S5. Input the target shape into the trained reverse reasoning model to generate the optimal design variables for the soft gripper; verify the optimal soft structure under the optimal design variables;

[0013] S6. Prepare and test the optimal software structure.

[0014] Preferably, the soft gripper includes a flexible base layer for stretching and retraction and a strain limiting layer with fixed stiffness disposed above the flexible base layer; the portion of the flexible base layer with the strain limiting layer forms the core gripping part, and the portions with stretching cutouts on both sides are the side stretching parts; the strain limiting layer includes multiple limiting strips.

[0015] As a preferred option, the design variables include: the overall length of the substrate, the length and width of the core gripping part, the horizontal angle between the core gripping part and the two side stretching parts, the stretching displacement, the thickness of the flexible substrate, the thickness of the strain limiting layer, and the bonding position of the strain limiting layer.

[0016] Preferably, the horizontal angle between the core gripping part and the two side stretching parts is defined as θ, and θ satisfies the following constraint:

[0017]

[0018] Where H is the overall length of the base, h is the length of the core gripping part, and L is the width of the core gripping part.

[0019] Preferably, the strain confinement layer consists of five confinement strips placed side by side; the width of the confinement strips and the spacing between them are... express, The following constraints must be met:

[0020]

[0021] Where h is the length of the core grabbing section.

[0022] Preferably, S2 includes: dividing the curve formed by the centerline contour point cloud of the deformed model into four groups of monotonic curves with different slopes; sorting the coordinate points according to the monotonic direction of each monotonic curve and then merging all the coordinate points; restoring the order of each coordinate point relative to the model; calculating the Euclidean distance between adjacent points after sorting and accumulating the Euclidean distances to obtain an approximate cumulative arc length value of each point relative to the starting point; mapping the two-dimensional coordinate point (x, y) to the arc length parameter domain, and performing linear interpolation on the x and y coordinates with respect to the arc length parameter s to obtain uniformly distributed standardized contour points as the new standardized centerline contour point cloud.

[0023] Preferably, S4 includes:

[0024] The forward proxy model and the backward inference model were trained using design variables and a contour point cloud dataset.

[0025] The mean squared error is defined as the loss function of the positive surrogate model;

[0026] During the training of the reverse inference model, a forward proxy model is used for auxiliary training.

[0027] As a preferred option, S4 also includes:

[0028] The contour point cloud corresponding to the design variables and the real point cloud are reconstructed based on the forward surrogate model to calculate the contour point cloud error. The contour point cloud error and the design variable error are combined into a joint error, and the inverse reasoning model is iterated from the reverse to the forward direction until convergence.

[0029] As a preferred embodiment, S6 includes:

[0030] S61. Provide a flexible substrate layer so that the soft gripper can be stretched or retracted along the length of the flexible substrate.

[0031] S62. A strain-limiting layer with fixed stiffness on a flexible substrate; the strain-limiting layer becomes unstable when stretched and spontaneously bends into a three-dimensional configuration, thereby achieving stable encapsulation and gripping of various objects.

[0032] Preferably, S62 includes:

[0033] S621. Place the prepared silicone into an inner groove mold that matches the shape of the strain limiting layer, and heat-cur it.

[0034] S622. After the preset curing time, soft silicone with porous properties is introduced into the mold, and chemical adhesive is used to achieve a tight bond between the two different silicones, thereby completing the preparation of the soft gripper.

[0035] Compared with the prior art, the beneficial effects of the present invention are reflected in:

[0036] 1. Unlike the multi-point grasping technology of traditional technology, the present invention adopts a three-dimensional structural deformation mechanism, which makes the soft gripper and the grasped object present an enveloping state, increasing the effective contact area and improving the grasping efficiency.

[0037] 2. Unlike the experience-based design technology of traditional technologies, this invention adopts a reverse design technology based on deep learning, which enables fast and efficient structural design for objects of arbitrary shapes.

[0038] 3. This invention provides a three-dimensional structural deformation method based on stretching drive, wherein the first layer is a porous flexible material and the second layer is a non-stretchable material. Due to the deformation redundancy of the porous structure and the large deformation characteristics of the flexible material, coupled with the deformation limitation of the second layer, the overall structure exhibits a three-dimensional deformation mode under stretching drive, realizing a wrapping gripping of objects.

[0039] 4. This invention provides a deep learning-based intelligent structural optimization model. Forward prediction employs a data-driven approach, establishing a surrogate model from structural parameters to mechanical response, enabling rapid performance evaluation and extensive exploration of the design space. Reverse structural design follows a goal-oriented approach, combining intelligent optimization methods to develop a reverse design model that inversely derives optimal structural parameters from desired functions (such as grasping shapes). Together, these two approaches transform the traditional lengthy closed loop of "design-simulation-iteration" into a highly efficient and intelligent "prediction-optimization" process, achieving a paradigm shift from experience-based trial and error to model-driven design. Attached Figure Description

[0040] Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention;

[0041] Figure 2 This is a schematic diagram of the soft gripper structure of Embodiment 1 of the present invention;

[0042] Figure 3 This is a side view of the soft gripper of Embodiment 1 of the present invention;

[0043] Figure 4 This is a schematic diagram of the design variables for the soft gripper in Embodiment 1 of the present invention;

[0044] Figure 5 This is a schematic diagram of the design variables for the soft gripper in Embodiment 1 of the present invention;

[0045] Figure 6 This is a schematic diagram of the point cloud data processing flow according to Embodiment 1 of the present invention;

[0046] Figure 7 This is a schematic diagram of the deformation of the soft gripper in Embodiment 1 of the present invention;

[0047] Figure 8 This is a schematic diagram of the mold for preparing the soft gripper according to Embodiment 2 of the present invention. Detailed Implementation

[0048] To make the technical means, inventive features, objectives, and effects of the invention readily understandable, the invention is further described below with reference to specific illustrations. However, the invention is not limited to the embodiments described below.

[0049] It should be noted that the structures, proportions, sizes, etc., illustrated in the accompanying drawings of this specification are only used to complement the content disclosed in the specification for those skilled in the art to understand and read, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0050] Example 1:

[0051] like Figure 1 As shown, a parametric design and intelligent optimization method for a tension-driven soft gripper is proposed. First, key geometric parameters are extracted and a parametric soft structure is randomly sampled. Numerical simulation is then performed using the finite element simulation software ABAQUS, and deformation information of the structure under different tensile displacements is extracted. Next, a neural network model is proposed to construct the physical mapping relationship between geometric parameters and deformation information. Finally, using a specified three-dimensional configuration as the target, the neural network model is used to generate a soft gripper that meets the requirements. Specifically, the method includes the following steps:

[0052] S1. Design a soft gripper including a core gripping part and two side stretching parts; the two side stretching parts are located at both ends of the core gripping part; use the parameters of each part of the soft gripper as design variables; randomly sample the combination of design variables within a specified range as parameter samples;

[0053] like Figure 2 , Figure 3 As shown, the soft gripper is based on a flexible porous structure, and this embodiment uses a silicone substrate. The soft gripper includes a flexible substrate layer 1 for stretching and retraction and a strain limiting layer 2 with fixed stiffness disposed above the flexible substrate layer 1. In this embodiment, the strain limiting layer 2 is an adhesive paper tape. The portion of the flexible substrate layer 1 with the strain limiting layer 2 forms the core gripping part, and the portions with stretching cutouts on both sides are the side stretching parts. The strain limiting layer includes multiple limiting strips.

[0054] like Figure 4 , Figure 5 As shown, the component parameters include: overall base length, length and width of the core gripping part, horizontal angle between the core gripping part and the two side tensioning parts, tensile displacement, flexible base thickness, strain-limiting layer thickness, and strain-limiting layer adhesion position. Using these component parameters as the design space, different parameter combinations are selected as samples.

[0055] The value ranges of parameters for each part are shown in Table 1:

[0056] Table 1: Design Variables and Value Ranges for Soft Grippers

[0057]

[0058] The horizontal angle between the core gripping section and the two side stretching sections is defined as θ, and θ satisfies the following constraint:

[0059]

[0060] Where H is the overall length of the base, h is the length of the core gripping part, and L is the width of the core gripping part.

[0061] The bonding positions of the strain-limiting layer and the flexible substrate layer are dynamically calculated based on the width and height of the core gripping part. The width of the limiting strips and their spacing are determined by... It means that among them The following constraints must be met:

[0062]

[0063] In addition to the range of values ​​mentioned above, some values ​​should also meet additional constraints to ensure that the model after modeling is physically feasible (such as lines not intersecting and paper tape not being suspended).

[0064] S2. Use ABAQUS to construct a geometric model of the soft gripper; obtain the node set displacement and initial coordinates of the geometric model; deform the soft gripper, and calculate the centerline contour point cloud of the deformed model based on the node set displacement and initial coordinates; Figure 7 This is a comparison diagram showing the initial and deformed states of the soft gripper.

[0065] A geometric model was built using the ABAQUS finite element simulation software through secondary development. Material properties were defined, meshes were created, and loads were set to perform static analysis. Based on the nodal set displacements and the initial model coordinates, the centerline contour point cloud of the deformed model was calculated. Quasi-static simulation was performed by automatically modeling, meshing, and applying tensile displacement boundary conditions using a Python script in conjunction with ABAQUS software. The centerline contour point cloud and global strain were extracted from the simulation results to create a dataset for subsequent model training.

[0066] like Figure 6 As shown, an additional standardization operation is performed on the centerline contour point cloud. The curve formed by the centerline contour point cloud is divided into four groups of monotonic curves with different slopes. The point clouds are sorted according to the monotonic direction of each curve. After sorting, the four segmented point clouds are merged to restore the order of each coordinate point relative to the model (from the beginning to the end of the curve relative to the cross-section of the model). The Euclidean distance between adjacent points after sorting is calculated, and these distances are accumulated to obtain the approximate cumulative arc length of each point relative to the starting point. The two-dimensional coordinate point (x, y) is mapped to the arc length parameter domain (s), and linear interpolation is performed on the x and y coordinates with respect to the arc length parameter s to obtain 30 uniformly distributed standardized contour points as the new standardized centerline contour point cloud.

[0067] S3. Combine parameter samples and deformed model centerline contour point cloud to construct a dataset of design variables and contour point cloud;

[0068] Combine the parameter samples generated in S1 with the deformed model centerline contour point cloud obtained in S2 to construct a dataset of design variables and contour point cloud.

[0069] The design variables and contour point cloud datasets were preprocessed by performing regularization and normalization on the parameter samples and shape data, respectively. The dataset was then divided into training, testing, and validation sets in a ratio of 7:2:1. Standard normalization was applied to the design variables, transforming the data into a normal distribution with a mean of 0 and a variance of 1. Shape data was scale-normalized, and the centerline contour point cloud was interpolated over the arc length parameter domain to obtain contour point clouds with the same number of points.

[0070] S4. Construct a forward surrogate model and a backward inference model, and train them jointly through a cycle consistency constraint: the forward surrogate model takes the design variables as input and outputs the corresponding contour point cloud; the backward inference model takes the target contour point cloud as input and outputs the corresponding design variables; during the training process, minimize the geometric error between the target contour point cloud and its reconstruction result through the "backward → forward" path to ensure that the backward inference result is physically realizable.

[0071] S41. Construct a forward surrogate model consisting of a multilayer perceptron and a convolutional neural network (input dimension 12, output dimension 2*30; hidden layers in the multilayer perceptron are [64, 128, 256]; the convolutional neural network uses four convolutional layers for downsampling, with kernel size and stride of [(2,2), (2,2), (2,2), (3,1)]). Construct a reverse prediction and inference model (input dimension 2*30, output dimension 12), with the target contour as input and design variables as output. The forward surrogate model assists in determining whether the generated design variables meet the target shape during the training process of the reverse inference model.

[0072] S42. Train the forward surrogate model and the backward inference model using the design variables and contour point cloud dataset. During the backward inference model training process, the forward surrogate model is used for auxiliary training. The model point cloud corresponding to the generated design variables is reconstructed using the forward surrogate model and the geometric error is calculated. The geometric error is combined with the original generation error to form the joint error. The backward model is iterated in a backward → forward loop until the backward inference model training converges.

[0073] Define the mean squared error (MSE) as the loss function of the positive surrogate model:

[0074]

[0075] in, To predict contour points, For true contour points, The number of points.

[0076] Define the weighted sum of design variable error and contour point cloud error as the loss function of the inverse prediction inference model:

[0077]

[0078] in, Design variables predicted based on the target shape. Design variables for real-world applications. These are the weighting coefficients. Number of design variables.

[0079] S5. Input the target shape into the trained reverse reasoning model to generate the optimal design variables for the soft gripper; verify the optimal soft structure under the optimal design variables;

[0080] The target design shape is input into the reverse inference model to generate the optimal design variables for the software structure; the optimal software structure is verified using the finite element simulation software ABAQUS to ensure that it can meet the target grasping shape.

[0081] S6. Prepare and test the optimal software structure.

[0082] Prepare the optimal software structure and conduct corresponding tests, including:

[0083] S61. A flexible substrate layer is provided, which can be stretched or retracted in a first extension direction. The first extension direction can be understood as the length direction of the flexible substrate.

[0084] S62. A high-stiffness strain-limiting layer is fixed on a flexible substrate. Utilizing the porous and stretchable properties of the flexible substrate and the non-stretchability of the strain-limiting layer, the layer becomes unstable under tension, spontaneously bending into a three-dimensional shape. This achieves stable encapsulation and gripping of various objects, including:

[0085] S621. Place the prepared silicone into an inner groove mold that matches the shape of the strain limiting layer, and heat-cur it to complete the component preparation.

[0086] S622. After heat curing for a certain period of time, the prepared soft silicone is poured into the mold, and chemical adhesive is used to achieve a tight bond between the two different silicones, thereby completing the preparation of the soft gripper.

[0087] The porous characteristic refers to the existence of different initial topological configurations in the structure, which affects the deformation path of the structure and is an important design variable for studying three-dimensional configurations. This gripper can grasp extremely soft or fragile objects, such as eggs, potato chips, and jelly. In addition, it also shows good performance in grasping highly irregularly shaped objects and sharp objects (such as cacti, pineapples, and durians).

[0088] Specifically, a pre-designed mold is printed using a 3D printer. A pre-mixed silicone PDMS layer is then poured into the mold to create a strain-constraining layer. Next, the mold is placed in a temperature-controlled oven at 60°C for 40 minutes. After 40 minutes, Dragon Skin-30 is prepared by mixing the two components at a 1:1 mass ratio, poured into the mold, and placed back into the oven at 60°C for 120 minutes. Finally, the integrated structure is removed from the mold using tweezers.

[0089] It should be noted that the shape of the flexible substrate in this embodiment is not specifically limited. Its porous structure 3 can be a symmetrical structure or an asymmetrical structure, or it can be different basic shapes (triangles, polygons, circles, etc.), which can further increase the diversity of the three-dimensional contour point cloud, thereby adapting to different objects being grasped.

[0090] Example 2:

[0091] For a food sorting task, it is necessary to safely grasp ripe strawberries (approximately 25mm in diameter, with a soft and fragile surface) and classify and package them. The desired outcome is that the soft gripper, after being driven, forms an approximately "U-shaped" envelope to achieve flexible wrapping without damaging the fruit. The target centerline contour is defined by 30 two-dimensional coordinate points (preprocessed according to the standardized method of this invention), denoted as... .

[0092] Based on a specified parameter range, 4000 sets of design variables are randomly generated. Each set of design variables contains... The range of design variable values ​​is shown in Table 1 above. All design variables must satisfy the following constraints to ensure that all lines do not intersect, and that the strain constraint layer has a reasonable coverage area, without overlap or gaps.

[0093]

[0094] Next, automatic modeling, mesh generation (element type C3D8H), application of displacement boundary conditions, and quasi-static simulation were performed in ABAQUS using Python scripts. The coordinates of the central axis nodes of the soft structure were extracted from the simulation results, and after topological sorting, arc length parameterization, and linear interpolation, 30 standardized contour data points were generated, while the global input strain was recorded. Based on the aforementioned Python scripts, batch data generation was implemented, and the data was normalized (4000 sets).

[0095] A forward surrogate model is constructed, consisting of a multilayer perceptron and a convolutional neural network (input dimension 12, output dimension 2*30; hidden layers in the multilayer perceptron are [64, 128, 256]; the convolutional neural network uses four convolutional layers for downsampling, with kernel size and stride of [(2,2), (2,2), (2,2), (3,1)]). The loss function of the forward surrogate model is defined as mean squared error.

[0096]

[0097] Set the learning rate to 1e-3 and train the positive surrogate model using the Adam optimizer until convergence.

[0098] Construct a reverse prediction inference model (input dimension 2*30, output dimension 12), with the target contour as the input. The output is the design variable. The loss function is defined as:

[0099]

[0100] The learning rate is set to 1e-3, and the inverse inference model is trained using the Adam optimizer until convergence. The previously described method for grasping the outline of the target strawberry is then applied. The trained inverse reasoning model is input to obtain the optimal structural design variables. Next, the design variables are inversely normalized using normalization parameters, and ABAQUS simulations are performed using a Python script to extract the numerical simulation structure and verify the model's correctness.

[0101] Furthermore, to verify the effectiveness of the model experimentally, a mold for preparing a soft gripper was designed based on the optimal design variables. The mold was then printed using a 3D printer, with polylactic acid as the printing material.

[0102] The prepared silicone is placed in an inner groove mold that matches the shape of the strain-limiting layer, and then thermosetting to complete the strain-limiting layer preparation. Next, prepared soft silicone is poured into the mold, and chemical adhesives are used to achieve a tight bond between the two different silicones, completing the soft gripper preparation. Figure 8 As shown.

[0103] The soft gripper was installed on the lead screw, and the gripping force and strawberry gripping performance were tested. It was found that it could completely wrap around the target shape and meet the gripping requirements.

Claims

1. A method for parametric design and intelligent optimization of stretch-driven soft gripper, characterized in that, Includes the following steps: S1. Design a soft gripper including a core gripping part and two side stretching parts; the two side stretching parts are located at both ends of the core gripping part; use the parameters of each part of the soft gripper as design variables; randomly sample the combination of design variables within a specified range as parameter samples; S2. Construct a geometric model for the soft gripper; obtain the node set displacements and initial coordinates of the geometric model; deform the soft gripper and calculate the centerline contour point cloud of the deformed model based on the node set displacements and initial coordinates; S3. Combine parameter samples and deformed model centerline contour point cloud to construct a dataset of design variables and contour point cloud; S4. Construct and train a forward surrogate model for predicting the target shape contour point cloud based on design variables and a backward inference model for inferring design variables based on the target shape contour point cloud. S5. Input the target shape into the trained reverse reasoning model to generate the optimal design variables for the soft gripper; verify the optimal soft structure under the optimal design variables; S6. Prepare and test the optimal software structure.

2. The parameterized design and intelligent optimization method of a stretch-driven soft gripper according to claim 1, wherein, The soft gripper includes a flexible base layer for stretching and retraction and a strain limiting layer with fixed stiffness located above the flexible base layer; the portion of the flexible base layer with the strain limiting layer forms the core gripping part, and the portions with stretching cutouts on both sides are the side stretching parts; the strain limiting layer includes multiple limiting strips.

3. The parametric design and intelligent optimization method for a stretch-driven soft gripper according to claim 2, characterized in that, Design variables include: overall substrate length, length and width of the core gripping part, horizontal angle between the core gripping part and the two side stretching parts, stretching displacement, thickness of the flexible substrate, thickness of the strain limiting layer, and adhesive position of the strain limiting layer.

4. The parametric design and intelligent optimization method for a stretch-driven soft gripper according to claim 3, characterized in that, The horizontal angle between the core gripping section and the two side stretching sections is defined as θ, and θ satisfies the following constraint: ； Where H is the overall length of the base, h is the length of the core gripping part, and L is the width of the core gripping part.

5. The parametric design and intelligent optimization method for a stretch-driven soft gripper according to claim 3, characterized in that, The strain-constraining layer consists of five confinement bars placed side by side; the width of the confinement bars and the spacing between them are... express, The following constraints must be met: ； Where h is the length of the core grabbing section.

6. The parametric design and intelligent optimization method for a stretch-driven soft gripper according to claim 1, characterized in that, S2 include: The curve formed by the point cloud of the centerline contour of the deformed model is divided into four groups of monotonic curves with different slopes; the coordinate points are sorted according to the monotonic direction of each monotonic curve and then all coordinate points are merged; the order of each coordinate point relative to the model is restored. Calculate the Euclidean distance between adjacent points after sorting, and accumulate the Euclidean distances to obtain the approximate cumulative arc length of each point relative to the starting point; map the two-dimensional coordinate point (x,y) to the arc length parameter domain, and perform linear interpolation on the x and y coordinates with respect to the arc length parameter s to obtain uniformly distributed standardized contour points as the new standardized centerline contour point cloud.

7. The parametric design and intelligent optimization method for a stretch-driven soft gripper according to claim 1, characterized in that, S4 include: The forward proxy model and the backward inference model were trained using design variables and a contour point cloud dataset. The mean squared error is defined as the loss function of the positive surrogate model; During the training of the reverse inference model, a forward proxy model is used for auxiliary training.

8. The parametric design and intelligent optimization method for a stretch-driven soft gripper according to claim 7, characterized in that, S4 also includes: The contour point cloud corresponding to the design variables and the real point cloud are reconstructed based on the forward surrogate model to calculate the contour point cloud error. The contour point cloud error and the design variable error are combined into a joint error, and the inverse reasoning model is iterated from the reverse to the forward direction until convergence.

9. The parametric design and intelligent optimization method for a stretch-driven soft gripper according to claim 1, characterized in that, S6 include: S61. Provide a flexible substrate layer so that the soft gripper can be stretched or retracted along the length of the flexible substrate. S62. A strain-limiting layer with fixed stiffness on a flexible substrate; the strain-limiting layer becomes unstable when stretched and spontaneously bends into a three-dimensional configuration, thereby achieving stable encapsulation and gripping of various objects.

10. The parametric design and intelligent optimization method for a stretch-driven soft gripper according to claim 9, characterized in that, S62 includes: S621. Place the prepared silicone into an inner groove mold that matches the shape of the strain limiting layer, and heat-cur it. S622. After the preset curing time, soft silicone with porous properties is introduced into the mold, and chemical adhesive is used to achieve a tight bond between the two different silicones, thereby completing the preparation of the soft gripper.

Images